EvacuatioNZ
Comparison with EvacuatioNZ
EvacuatioNZ is a network evacuation model by M. Spearpoint. It uses what its author calls “the well established network method” [7, manuscript p. 3]; its sources are listed under network models. This model follows the same approach but is not equivalent to it: several algorithms differ, and no EvacuatioNZ runs were made for this comparison. No licensed copy of EvacuatioNZ was used. EvacuatioNZ has an online user manual [12]; the table rests on published documents, which describe different versions of EvacuatioNZ: 1.2 [5], 2.3 [8], 2.8 [10], 2.11 [2], 2.11.2 [3, p. 3] and 2.14 [6]. Each row names the version where it matters.
| Aspect | EvacuatioNZ | jupedsim_network |
|---|---|---|
| Speed in a node | SFPE $S = k(1-0.266D)$, $D \ge 0.54$ [2, §2.1.3; 3, p. 15] | same |
| Stair speed | $k = 51.8\sqrt{T/R}$ m/min [2, §2.1.3, Eq. 2.1], “the FEDG approach” [3, p. 17], from the FEDG’s $k_t = 51.8\,(G/R)^{0.5}$ with tread going $G$ [18, p. 223] | same formula |
| Door flow | $F_s = 1.33$ default, boundary layer 0.15 m [2, §2.2; 3, p. 3]; door leaves and closers [2, §2.2; 3, p. 14] | $F_s = 1.3$ default (SFPE), configurable per link; no leaves or closers |
| Stairs | stair node with an enz_stairs connection; $F_s = 1.09$ from SFPE Table 59.5 in the exercise [3, p. 18]; landings as separate nodes [3, p. 20] | stair node with area and $k(R,T)$; link flow $k/(4a)$ computed from the geometry |
| Node capacity | hard maximum node density, default 2.75 m⁻² [3, p. 3]: an agent passes into a node only if its density is below the maximum [8, p. 165; 10, p. 111] | hard max_density plus a linear supply reduction above 1.88 m⁻² |
| Agent size | one density for all agents. In v1.2 the user lowered the local door queue density, calibrated against EXIT89 and Simulex, to represent disabled occupants; the model cannot represent body sizes [5, pp. 25, 54, 62, 69] | area factor per agent, counted in density and free space |
| Update order | sequential by default (enz_sequential); random orders are optional [3, pp. 3, 42] | synchronous; simultaneous arrivals at one queue in population order |
| Queue near the door | “congested node” algorithm shortens the walk of agents who join a queue: 22.5 s with it, 31.5 s without, 30.5 s by hand [3, pp. 15–16] | not modelled; agents walk the full link length at the node speed |
| Merging | 50 : 50 by default; <Merge> weights 1 : 3 gave 2.8 : 1 [3, p. 35] | weights, interleaved by virtual time; 1 : 3 gives about 3 : 1 |
| Counterflow | v2.14 applies half the effective door width under counterflow [6, p. 8]; the 2022 exercise is still marked “To do” [3, p. 43]. In Arnott et al.’s trials EvacuatioNZ gave 0.96 persons/s per direction at a 1.72 m door [6, Table 1, p. 10]. The published sources do not state when v2.14 applies the rule or how it forms the half width | both directions share one door; while both queue they pass together $g\,C$ with $g$ = 0.84–1.0 (stairs 0.75–1.0), split by queue share with at least 17 % per direction; default "bounding". At 50/50, "estimate" passes $C/2$ each way, the halving Arnott et al. report for v2.14 (Counterflow). The two rules coincide only there |
| Routes | minimum or maximum distance, minimum number of nodes, random, specified exit, required connection [2, §5.1–5.2]; a random choice when two or more required or exit-sign connections leave a node [3, pp. 53–54]; an optional least populated connection subtype of minimum distance, used “to get a close to 50:50 split” [2, §2.6]; preferred paths and probabilistic assignment [8, p. 165]. How minimum distance breaks an exact tie is not documented | shortest distance by default, shortest free-flow reference time with routing_policy="shortest_free_flow_time", or a fixed target, no reassessment; agents alternate between equally short routes (Networks), a rule of this model, not taken from EvacuatioNZ |
| Pre-movement | fixed, uniform, normal, log-normal [2, §3.1], triangular [2, §3.2] and Weibull, truncatable [10, p. 110] | the same six types, truncated at 0 by default |
| Lighting | lighting in v1.2 [5, pp. 40–42, 89–92] | no |
| Smoke | v2.3 could not close an escape route because of fire or smoke spread [8, p. 165] | no |
| Verification | travel speed, door and stair flow including the IMO-style cases [2, §2.1–2.3]; Fire Engineering Design Guide example, 179 s against 186 s by hand [2, §2.4]; SFPE 9-storey example, 1871 s against 1524 s by hand [2, §2.5]; SFPE human behaviour guide examples, 221.5 s against 246 s and 283 s against 300 s [2, §2.6]; exercises against hand calculations [7] | IMO tests 1, 3 (adapted) and 4, plus hand calculations; the cases of the EvacuatioNZ verification report rebuilt and compared with its published results, including the SFPE 9-storey example with occupants and exit doors taken from the SFPE source and tied stairs split by alternation (Verification); the Station nightclub network of [8] is rebuilt and compared with the published v2.3 results (The Station nightclub) |
| Validation | Jean Talon drill: 827 s (mean of 100 runs) against 863 s observed, with the pre-movement spread chosen to fit the drill and many inputs hypothesised [5, pp. 74, 82–88]; 21-storey office drill in Australia, one recorded value of 1800 s against 1730–1830 s simulated [5, pp. 103–105]; v2.8 against 14 stair evacuations, 95th-percentile times within ±17 % except one building up to 26 % faster [10, pp. 115–116]; lecture rooms [2, §6.2; 9]. The author of the Station nightclub study states that it would be inappropriate to claim that the work validated EvacuatioNZ [8, p. 180] | none |
| Network resolution | on a hypothetical building, a network of 61 nodes and one of 496 nodes gave relatively small differences [5, pp. 53, 63–64] | not studied |
| Licence | provided under the licence shipped with the program [3, p. 3] | open source, LGPL-3.0 |
This model’s congested walking test (200 agents at 2 m⁻², 20 m) uses the case of the congested-node exercise [3, p. 15].
Last updated on